I am trying to examine the effects of studentification on private residents in a studentified area, either it is positive or negative (which is coded as 1 or 0 respectively) as the dependent variable.
The independent variables are effects of studentification (across literatures) on 5-likert scale.
The question is am I to also dichotomies the likert scale responses from (strongly disagree, disagree, neutral, agree and disagree) to (1: positive, 0: negative)?
Dear Matthew Femi Adeboye
A very important question, thank you for asking
We do not recommend using logistic regression in your case to determine the effect in this way
If you keep a Likert of five points, you can make a comparison between the two cases of the dependent variable
Using the median test, and other non-parametric tests such as Mann-Whitney
A statistically significant difference is an indication of the presence of an effect.
Dear Matthew Femi Adeboye
If you want to apply logistic regression, your variables must be qualitative with a score of 2 * 2 To be able to calculate the relative risk factor and odds ratio.
There you have an example where we used logistic regression
Benefit from commenting on the results and displaying the tables
To understand the idea more..
Article Cleft lip and palate risk factors among otorhinolaryngology ...
I have 2 comments.
1) There are no distributional assumptions for the explanatory variables.
2) Coarsening variables in the way you suggest throws away information and therefore tends to reduce power.
So no, you do not have to dichotomize your explanatory variable(s). But you should be mindful of rules of thumb about the number of events per explanatory variable degrees of freedom that are needed to reduce the likelihood of overfitting your model. See the datamethods.org checklist, for example, and search for 20:1.
https://discourse.datamethods.org/t/author-checklist/3407
By the way, that checklist also has a section on coarsening variables. Look under "Inefficient use of continuous variables". Your Likert-type items are not continuous, of course, but the general principle still applies.
HTH.
Bruce Weaver
Sir, when i run the logistic regression analysis without dichotomizing (i left the 5-likert scale responses as coded from 1 to 5), in the "the variable in the equation" table, all the independent variables' sig all have not less than .998 p-values. So this confuses me.
Is it normal or where am i getting it wrong?
Matthew Femi Adeboye, it is impossible to say without seeing your command syntax. It would also be helpful to see the output. One way things could be going wrong is failure to treat your 5-point item as categorical. See the way the SES variable is treated on this page for an example:
https://stats.oarc.ucla.edu/spss/output/logistic-regression/
And notice in the table of coefficients that there is an overall Wald test (with 2 df) for SES. In your case, you should see an overall Wald test with 4 df for your 5-point item.
HTH.
As you have an ordinal dependent variable measured on a 5-point Likert scale (or any ordinal scale), you might want to consider using an ordinal logistic regression model.
Ordinal logistic regression is suitable for modeling the relationship between one or more independent variables and an ordinal outcome with more than two categories and you can run this in SPSS.
Md. Sabit Al-Saba Rion, please look again at the first sentence of the original question:
I am trying to examine the effects of studentification on private residents in a studentified area, either it is positive or negative (which is coded as 1 or 0 respectively) as the dependent variable.
The DV is dichotomous, not ordinal with 5 levels. The Likert-type variables are explanatory variables, as I read things.
Bruce Weaver Sir, thank you for correcting me. I was just suggesting another way of analyzing this situation in SPSS. Because if one makes this a dichotomous variable then it would lead to losing insights from the data. And the cut-off point should be decided carefully to make this a dichotomous variable.
Md. Sabit Al-Saba Rion, I still see no explicit statement in the original question that the DV originally had more than 2 levels and was dichotomised so that binary logistic regression could be used. But as I look at it again, there is a hint that it may have been done. I am referring to this part:
The question is am I to also dichotomies the likert scale responses from (strongly disagree, disagree, neutral, agree and disagree) to (1: positive, 0: negative)?
The highlighted word also suggests that this may have been done for the DV. And if it was, then I agree with you that it should not have been done, and that ordinal logistic regression using the original DV would be a better choice.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression